Intelligent wheelchair health monitoring system based on multi-modal sensor fusion

Abstract

The invention is applicable to the technical field of wheelchairs, and provides an intelligent wheelchair health monitoring system based on multi-modal sensor fusion, comprising: a sensor module comprising a reflective PPG sensor, an ECG electrode array, and a six-axis IMU sensor; the signal processing module is configured with a wheelchair exclusive motion compensation algorithm, constructs a wheelchair vibration noise model based on the IMU data, and separates effective physiological signals from PPG and ECG through adaptive Kalman filtering; the multi-modal blood pressure prediction module is used for calculating a blood pressure value through a machine learning model in combination with PPG waveform characteristics, ECG-PTT time differences and thoracic impedance change characteristics; the low-power-consumption management module dynamically controls a high-power-consumption sensor switch according to the wheelchair motion state; a user information collection module; by adopting a wheelchair exclusive motion compensation algorithm, bad data are eliminated, accurate data are obtained, and the heart rate and the blood pressure of the user are accurately measured.

Description

Technical Field

[0001] This invention relates to the field of health monitoring technology, and more specifically, to an intelligent wheelchair health monitoring system based on multimodal sensor fusion. Background Technology

[0002] Portable health monitoring devices, especially physiological parameter acquisition systems integrated into smart wheelchairs, have gradually become key assistive tools in chronic disease management, elderly care, and rehabilitation medicine in recent years. With the increasing aging population and rising demand for home-based healthcare, how to acquire users' health data in real-time, accurately, and with low power consumption in mobile scenarios has become an important research direction in the design of health monitoring systems.

[0003] In existing technologies, health monitoring systems typically rely on single-modal physiological signal acquisition methods, such as using only electrocardiogram (ECG) or photoplethysmography (PPG) sensors to estimate heart rate and blood pressure. While these methods offer a degree of accuracy in static environments, they become susceptible to signal distortion during movement, particularly in wheelchair-bound scenarios, due to environmental vibrations, changes in posture, and unstable sensor contact. This leads to decreased monitoring accuracy, frequent misjudgments, and even system failure.

[0004] For example, patent CN103565429A discloses a method for improving heart rate measurement accuracy by combining PPG and ECG signals. This method achieves relatively reliable detection results by selecting the optimal signal from multiple signal channels, thus improving signal quality. However, this solution still fails to address the problem of actively suppressing dynamic interference signals, nor does it propose an effective strategy for power consumption management during motion. Especially in long-term continuous monitoring tasks, the continuous acquisition of high-power physiological signals will lead to rapid battery depletion, making it difficult to meet the actual battery life requirements of wheelchair users.

[0005] Therefore, based on the above problems, this application proposes an intelligent wheelchair health monitoring system based on multimodal sensor fusion. Summary of the Invention

[0006] In view of the shortcomings of the existing technology, the purpose of this invention is to provide an intelligent wheelchair health monitoring system based on multimodal sensor fusion.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A smart wheelchair health monitoring system based on multimodal sensor fusion includes:

[0009] The sensor module includes:

[0010] A reflective PPG sensor is embedded in the back of the wheelchair, positioned 5cm above the center line of the backrest. It uses an 850nm wavelength light source with an incident angle of 30°±5° to monitor back vascular signals.

[0011] The ECG electrode array is located in the inner recessed area of ​​the armrest and the backrest, and is used to collect dual-lead ECG signals by conforming to the thenar eminence of the user's palm. It also has a capacitive contact status detection function.

[0012] A six-axis IMU sensor, mounted on the wheelchair base, is used to collect wheelchair vibration data in the 0.5Hz to 5Hz frequency band;

[0013] The signal processing module is equipped with a wheelchair-specific motion compensation algorithm, which constructs a wheelchair vibration noise model based on the IMU data and separates effective physiological signals from PPG and ECG through adaptive Kalman filtering.

[0014] The multimodal blood pressure prediction module combines PPG waveform features, ECG-PTT time difference, and chest impedance change features to calculate blood pressure values ​​using a machine learning model.

[0015] The low-power management module dynamically controls the switching of high-power sensors based on the wheelchair's movement status;

[0016] The user information collection module is used to initiate PPG data collection when the user is seated, initiate ECG data collection when the wheelchair is stationary, and integrate multimodal data to output heart rate and blood pressure information. In abnormal situations, it triggers local vibration alerts.

[0017] The present invention is further configured such that the reflective PPG sensor is positioned away from the scapula, acquires signals through the skin of the back to avoid interference from clothing, and improves the signal-to-noise ratio.

[0018] The present invention is further configured such that: the wheelchair-specific motion compensation algorithm is used to filter non-physiological noise signals introduced by wheelchair vibration, collision, bumps, etc., to perform frequency domain modeling of vibration signals, and to achieve real-time noise filtering through an adaptive Kalman filter.

[0019] The present invention is further configured such that the vibration signal is concentrated in the range of 0.5~5Hz, wherein the vibration source includes back vibration, handrail vibration, and collision or downhill.

[0020] The present invention is further configured such that: the smoothing coefficient is only involved in the calculation of the compensation algorithm when the wheelchair is in motion; when the IMU sensor detects that the wheelchair is stationary and stable, the system skips all historical data and motion interference modeling, and directly outputs the current PPG and ECG raw signals as health monitoring values.

[0021] During movement, a wheelchair-specific motion compensation algorithm is used for processing.

[0022] The present invention is further configured such that the wheelchair-specific motion compensation algorithm includes the following steps:

[0023] S1. Data from the IMU's three-axis acceleration system. Perform exponentially weighted smoothing:

[0024]

[0025] Current acceleration;

[0026] Smoothing coefficient;

[0027] S2. Construct the Kalman filter model and set the state equation and observation equation:

[0028] The state equation is as follows:

[0029]

[0030] The observation equation is:

[0031]

[0032] Among them, For the target signal state, To interfere with the input, These are observations, including ECG and PGG data. Noise term;

[0033] S3. Using the Kalman gain correction estimate:

[0034]

[0035] S4. Perform outlier identification and processing. If the variation between consecutive peaks is greater than twice the standard deviation, it can be judged as an anomaly.

[0036] The present invention is further configured such that: the signal processing module also includes a signal reliability screening mechanism, which calculates the wheelchair motion level based on the triaxial acceleration data collected by the IMU sensor, and accordingly eliminates interference values ​​in the physiological signal acquisition results. The motion level calculation formula is:

[0037]

[0038] in , , These represent the instantaneous acceleration values ​​of the wheelchair along the three axes.

[0039] The present invention is further configured such that: the interference value elimination process is performed by setting a threshold. and stable time ,when Exceeding the set threshold At the same time, the system automatically reduces the weight of PPG and ECG signals in data fusion and blood pressure prediction, and pauses some physiological signal processing to avoid misjudgment caused by motion interference.

[0040] when Below the threshold When, and the duration exceeds the set stabilization time. At that time, the system initiates dual-modal fusion processing and outputs real-time heart rate and blood pressure values.

[0041] The present invention is further configured such that: when the low-power management module detects wheelchair movement, it shuts down the high-power ECG sensor and only keeps the IMU and PPG working to save power.

[0042] The present invention is further configured such that: the abnormal state includes heart rate exceeding a set threshold, blood pressure continuously fluctuating abnormally, or signal disconnection, and the system can trigger a local vibration motor to provide a physical reminder.

[0043] In summary, this application includes at least one of the following beneficial technical effects:

[0044] 1. A complete multimodal physiological signal acquisition platform was constructed by comprehensively employing a reflective PPG sensor, an ECG dual-lead electrode array, and a six-axis IMU inertial sensor. By introducing an adaptive Kalman filter algorithm and combining it with wheelchair acceleration and attitude change data acquired by the IMU, a motion disturbance model was dynamically constructed to achieve real-time modeling and filtering of non-physiological disturbances in the PPG and ECG. Furthermore, the system further manages the weighted signals at different stages through a signal reliability screening mechanism, calculates the signal stability index based on the instantaneous motion level of the wheelchair, and adjusts the participation weight of each modality data in the fusion algorithm accordingly to ensure the reliability and stability of the final output data. This significantly improves the monitoring accuracy and anti-interference capability of key physiological indicators such as heart rate and blood pressure for wheelchair users in typical high-interference conditions such as bumpy outdoor roads, slow downhill driving, or sharp turns, thereby reducing the false alarm rate and increasing the application value of the system in real-world scenarios.

[0045] 2. A low-power management strategy based on wheelchair motion status is introduced into the system design. This strategy uses the IMU module to sense in real time whether the wheelchair is stationary or in motion, and dynamically controls the working status of each sensor accordingly: when the wheelchair is detected to be in a non-stationary state, the system automatically shuts down the high-power ECG module, retaining only the reflective PPG and IMU for simplified signal acquisition; only when the wheelchair is detected to be stationary for an extended period and the stability criterion is met, the ECG module is reactivated and the multimodal fusion mechanism is initiated. This dynamic scheduling method ensures the integrity of physiological data at critical nodes while effectively avoiding unnecessary energy waste. Compared with the traditional continuous full-time acquisition method, this method can significantly extend the continuous working time of the device after a single charge, reduce the frequency of device maintenance, and greatly improve the availability of the system in long-term scenarios such as home rehabilitation and outdoor activities for the elderly. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the system hardware layout of an intelligent wheelchair health monitoring system based on multimodal sensor fusion according to the present invention.

[0047] Figure 2 This is a flowchart of the algorithm for an intelligent wheelchair health monitoring system based on multimodal sensor fusion according to the present invention.

[0048] Figure 3 This is a flowchart of an intelligent wheelchair health monitoring method based on multimodal sensor fusion according to the present invention. Detailed Implementation

[0049] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0050] It should be noted that, unless otherwise specified, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0051] Please see Figure 1-3 The present invention provides the following technical solutions:

[0052] Example 1: An intelligent wheelchair health monitoring system based on multimodal sensor fusion includes a sensor module, a signal processing module, a multimodal blood pressure prediction module, a low-power management module, and a user information collection module. Through the collaborative work of multiple sensors and algorithm strategies customized for wheelchair scenarios, stable, efficient, and low-power monitoring of the user's physiological parameters such as heart rate and blood pressure is achieved.

[0053] The sensor module includes a reflective PPG sensor, an ECG electrode array, and a six-axis IMU sensor.

[0054] A reflective PPG sensor is embedded in the wheelchair backrest, positioned 5cm above the center line to avoid the scapular region and reduce light scattering caused by skeletal obstruction. This placement also minimizes reliance on the user's posture and position. The light source used has a wavelength of 850nm, belonging to the near-infrared band, which has strong tissue penetration capabilities. The incident angle is designed at 30°±5° to maintain stable reflection intensity while minimizing interference from clothing, backrest structure, and other factors on the optical signal. The sensor acquires the volumetric pulse rate (PPG) signal from the back vessels, providing basic data for subsequent heart rate and blood pressure estimation.

[0055] Two sets of ECG electrode arrays are respectively located in the recessed areas on the inner side of the armrests and on the backrest of the wheelchair. The electrode areas are designed to conform to the thenar eminence of the human hand, ensuring a large and stable contact area. This structure is used for acquiring dual-lead electrocardiograms (ECG).

[0056] A six-axis IMU sensor is mounted on the bottom bracket of the wheelchair, using a module with a three-axis accelerometer and a three-axis gyroscope. The IMU is used to collect vibration and dynamic acceleration changes of the wheelchair in the 0.5Hz to 5Hz frequency range to help the system determine whether the wheelchair is in motion.

[0057] The signal processing module is equipped with a wheelchair-specific motion compensation algorithm, which constructs a wheelchair vibration noise model based on the IMU data and separates effective physiological signals from PPG and ECG through adaptive Kalman filtering.

[0058] The multimodal blood pressure prediction module combines PPG waveform features, ECG-PTT time difference, and chest impedance change features to calculate blood pressure values ​​using a machine learning model.

[0059] Specifically, this module comprehensively uses multiple physiological features such as PPG waveform, ECG-PTT (pulse conduction time), and chest impedance characteristics, and uses machine learning models (such as random forest) to predict blood pressure values.

[0060] When a user first uses the system, it collects 10 minutes of baseline data (PPG + ECG - PTT) and combines this data with the user's historical information, such as age, gender, and baseline blood pressure records, to calibrate the model. Subsequent data collection involves continuously updating predicted values ​​using a sliding window approach. Multimodal data fusion effectively improves the prediction accuracy of systolic and diastolic blood pressure, especially in wheelchair vibration scenarios, where it dynamically adjusts the weights of each channel to enhance robustness.

[0061] The low-power management module dynamically controls the switching of high-power sensors based on the wheelchair's movement status. When the low-power management module detects wheelchair movement, it shuts down the high-power ECG sensor, keeping only the IMU and PPG working to save power.

[0062] The user information collection module is used to initiate PPG data acquisition when the user is seated and ECG data acquisition when the wheelchair is stationary. It integrates multimodal data to output heart rate and blood pressure information and triggers a local vibration alert in abnormal situations. These abnormal situations include heart rate exceeding a set threshold, persistent abnormal blood pressure fluctuations, or signal disconnection. Specifically, the ECG records cardiac electrical activity through electrodes; its R wave corresponds to ventricular depolarization and serves as a marker for heart rate detection.

[0063] In actual use, once the user is seated, the PPG sensor on the backrest automatically starts to collect back vascular signals. When the IMU detects that the wheelchair is stationary, the ECG electrode is activated. The dual-modal data is fused through Kalman filtering to output the real-time heart rate. In case of abnormality, a local vibration alert is triggered.

[0064] The wheelchair-specific motion compensation algorithm is used to filter non-physiological noise signals introduced by wheelchair vibration, collision, bumps, etc. It performs frequency domain modeling of vibration signals and achieves real-time noise filtering through an adaptive Kalman filter, thereby correcting the signals input from ECG and PPG to obtain the corrected signals.

[0065] The vibration signal is concentrated in the range of 0.5~5Hz, and the vibration sources include back vibration, handrail vibration, and collision or downhill.

[0066] The smoothing coefficient is only used in the compensation algorithm calculation when the wheelchair is in motion; when the IMU sensor detects that the wheelchair is stationary and stable, the system skips all historical data and motion interference modeling, and directly outputs the current PPG and ECG raw signals as health monitoring values.

[0067] During movement, a wheelchair-specific motion compensation algorithm is used for processing.

[0068] The wheelchair-specific motion compensation algorithm includes the following steps:

[0069] S1. Data from the IMU's three-axis acceleration system. Perform exponentially weighted smoothing:

[0070]

[0071] Current acceleration;

[0072] Smoothing coefficient;

[0073] S2. Construct the Kalman filter model and set the state equation and observation equation:

[0074] The state equation is as follows:

[0075]

[0076] The observation equation is:

[0077]

[0078] Among them, The target signal status (ECG or PPG signal that has been cleared of interference). To interfere with the input, The actual observed value (the original detected ECG or PPG signal). Noise term;

[0079] S3. Using the Kalman gain correction estimate:

[0080]

[0081] S4. Perform outlier identification and processing. If the variation between consecutive peaks is greater than twice the standard deviation, it can be judged as an anomaly and will not be included in the subsequent processing range.

[0082] In the specific signal processing, the wheelchair-specific motion compensation algorithm not only filters and denoises the PPG and ECG signals, but also performs signal quality assessment and data filtering. When the IMU sensor detects significant vibration in the wheelchair and the acceleration value fluctuation exceeds a preset threshold (e.g., 0.3g), the system determines that the acquired signal during that period is significantly affected by motion interference and contains strong non-physiological noise interference. The system automatically marks the ECG and PPG signals within the current time window as "low confidence" and excludes them from the heart rate and blood pressure calculation process.

[0083] Among them, the smoothing coefficient Adaptive selection is possible. Generally, when the smoothing coefficient is set to 1, only the current value is considered, resulting in no smoothing and retaining all noise without filtering. When the smoothing coefficient is set to 0, only historical values ​​are considered, resulting in no updates and system lockup. In this embodiment, the smoothing coefficient is automatically adjusted using an exponential function decay method.

[0084]

[0085] in, Set to 0.7, Set to 0.3,

[0086] k is the sensitivity adjustment coefficient, which is preferably 1.5 in this embodiment.

[0087] The above-mentioned smoothness coefficient design ensures that during vigorous wheelchair movement, The signal tends to This can make the signal more stable.

[0088] Example 2: Although Example 1 already achieved multimodal health signal acquisition and processing, in real-world scenarios involving prolonged continuous use, complex road conditions, or multi-user switching, the surge in data volume may lead to computational delays or resource conflicts. Therefore, Example 2 further introduces a signal reliability filtering mechanism based on Example 1. This signal reliability filtering mechanism calculates the wheelchair motion level based on triaxial acceleration data acquired by the IMU sensor and accordingly eliminates interference values ​​in the physiological signal acquisition results. The motion level calculation formula is:

[0089]

[0090] in , , These represent the instantaneous acceleration values ​​of the wheelchair along the three axes.

[0091] The interference elimination process involves setting a threshold. and stable time ,when Exceeding the set threshold At the same time, the system automatically reduces the weight of PPG and ECG signals in data fusion and blood pressure prediction, and pauses some physiological signal processing to avoid misjudgment caused by motion interference.

[0092] when Below the threshold When, and the duration exceeds the set stabilization time. At that time, the system initiates dual-modal fusion processing and outputs real-time heart rate and blood pressure values.

[0093] Example 3: A monitoring method for an intelligent wheelchair health monitoring system based on multimodal sensor fusion, applicable to real-time, low-power, and high-accuracy monitoring of key physiological parameters such as heart rate and blood pressure by wheelchair users during daily travel, rehabilitation training, or care. The method specifically includes the following steps:

[0094] Step S1: Initialization phase;

[0095] After system startup, the first step is to initialize each sensor module, including the PPG sensor, ECG electrode array, IMU module, signal processor, storage module, and communication interface module. This stage also includes the following sub-processes:

[0096] The system confirms the connection status and initial level of each sensor;

[0097] Set default parameter thresholds, including motion thresholds. Heart rate / blood pressure alarm thresholds, Kalman filter initial values, etc.;

[0098] Collect basic user information, such as age, gender, height, weight, and historical blood pressure data, for use in subsequent model initialization.

[0099] Step S2: User state recognition and sensory activation;

[0100] Determine if the user is firmly seated using IMU three-axis acceleration:

[0101] When the wheelchair is stationary (acceleration value is stable), Below and maintain When the time exceeds 1 second, activate the PPG and ECG sensors;

[0102] If the system determines that the user is not sitting still or is in a state of obvious movement, it will only turn on the low-power IMU and PPG for pre-monitoring, turn off the high-power ECG array, and enter standby or pre-acquisition mode.

[0103] Step S3: Data Synchronization Acquisition;

[0104] After system activation, multi-channel synchronous acquisition of physiological data begins:

[0105] Reflective PPG acquisition of back volumetric pulse wave signals;

[0106] ECG dual-lead acquisition of electrocardiogram data;

[0107] The IMU collects motion and vibration interference signals;

[0108] If a chest impedance module is configured, respiratory or chest dynamic change data can also be collected;

[0109] All acquired signals are timestamped and transmitted to the signal processing module in a unified manner.

[0110] Step S4: Motion compensation and signal cleaning;

[0111] A vibration noise model was constructed based on IMU data, and motion compensation processing was performed on PPG and ECG signals. The steps are as follows:

[0112] S41. Data on IMU triaxial acceleration Perform exponentially weighted smoothing:

[0113]

[0114] Current acceleration;

[0115] Smoothing coefficient;

[0116] S42. Construct a Kalman filter model and set the state equation and observation equation:

[0117] The state equation is as follows:

[0118]

[0119] The observation equation is:

[0120]

[0121] Among them, For the target signal state, To interfere with the input, For the observed values, Noise term;

[0122] S43. Using the Kalman gain correction estimate:

[0123]

[0124] S44. Perform outlier identification and processing. If the variation between consecutive peaks is greater than twice the standard deviation, it can be judged as an anomaly.

[0125] Step S5: Perform multimodal blood pressure prediction;

[0126] Based on the cleaned signals, execute the blood pressure prediction process:

[0127] Extract PPG features;

[0128] Calculate the pulse conduction time (PTT) between the ECG and PPG waveforms.

[0129] If chest impedance is collected, it is used as an auxiliary parameter input;

[0130] All features are used as vector inputs to a pre-trained machine learning model;

[0131] Output the current estimated systolic and diastolic blood pressure values;

[0132] A sliding window strategy is used to continuously update the predicted values ​​to avoid misjudgments caused by instantaneous fluctuations.

[0133] Step S6: Trustworthiness screening and power consumption control;

[0134] Combined with IMU motion level index Dynamic reliability regulation is implemented based on physiological signal stability scores.

[0135] like > The signal quality deteriorates, the system reduces the fusion weight, and pauses blood pressure updates;

[0136] like < And it remains stable, restoring PPG+ECG dual-modal fusion;

[0137] At the same time, the power consumption module shuts down high-energy-consuming components such as the ECG acquisition unit, retaining only the necessary modules;

[0138] The system enters energy-saving polling mode at night or when there are no significant fluctuations for a long period of time.

[0139] Step S7: Information feedback and anomaly alerts;

[0140] When the system outputs abnormal heart rate or blood pressure parameters (the system initially sets heart rate and blood pressure, typically to heart rate > 120 bpm and blood pressure > 160 / 100 mmHg), the following actions are triggered:

[0141] The local vibration module provides a slight vibration alert to the user or caregiver.

[0142] The device embodiments described above are merely illustrative and not all embodiments. For example, the division of units is only a logical functional division; in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the shown or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces. Indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms. All other embodiments obtained by those skilled in the art without inventive effort should fall within the scope of protection of this invention.

Claims

1. An intelligent wheelchair health monitoring system based on multimodal sensor fusion, characterized in that: include: The sensor module includes: A reflective PPG sensor is embedded in the back of the wheelchair, positioned 5cm above the center line of the backrest. It uses an 850nm wavelength light source with an incident angle of 30°±5° to monitor back vascular signals. The ECG electrode array is located in the inner recessed area of ​​the armrest and the backrest, and is used to collect dual-lead ECG signals by conforming to the thenar eminence of the user's palm. It also has a capacitive contact status detection function. A six-axis IMU sensor, mounted on the wheelchair base, is used to collect wheelchair vibration data in the 0.5Hz to 5Hz frequency band; The signal processing module is equipped with a wheelchair-specific motion compensation algorithm, which constructs a wheelchair vibration noise model based on the IMU data and separates effective physiological signals from PPG and ECG through adaptive Kalman filtering. The multimodal blood pressure prediction module combines PPG waveform features, ECG-PTT time difference, and chest impedance change features to calculate blood pressure values ​​using a machine learning model. The low-power management module dynamically controls the switching of high-power sensors based on the wheelchair's movement status; The user information collection module is used to initiate PPG data collection when the user is seated, initiate ECG data collection when the wheelchair is stationary, and integrate multimodal data to output heart rate and blood pressure information. In abnormal situations, it triggers local vibration alerts.

2. The intelligent wheelchair health monitoring system based on multimodal sensor fusion according to claim 1, characterized in that: The reflective PPG sensor is positioned away from the shoulder blades, acquiring signals through the skin of the back to avoid interference from clothing and improve the signal-to-noise ratio.

3. The intelligent wheelchair health monitoring system based on multimodal sensor fusion according to claim 1, characterized in that: The wheelchair-specific motion compensation algorithm is used to filter non-physiological noise signals introduced by wheelchair vibration, collision, bumps, etc., to perform frequency domain modeling of vibration signals, and to achieve real-time noise filtering through an adaptive Kalman filter.

4. The intelligent wheelchair health monitoring system based on multimodal sensor fusion according to claim 3, characterized in that: The vibration signal is concentrated in the range of 0.5~5Hz, and the vibration sources include back vibration, handrail vibration, and collision or downhill.

5. The intelligent wheelchair health monitoring system based on multimodal sensor fusion according to claim 3, characterized in that: The smoothing coefficient is only used in the compensation algorithm calculation when the wheelchair is in motion; when the IMU sensor detects that the wheelchair is stationary and stable, the system skips all historical data and motion interference modeling, and directly outputs the current PPG and ECG raw signals as health monitoring values. During movement, a wheelchair-specific motion compensation algorithm is used for processing.

6. The intelligent wheelchair health monitoring system based on multimodal sensor fusion according to claim 5, characterized in that: The wheelchair-specific motion compensation algorithm includes the following steps: S1. Data from the IMU's three-axis acceleration system. Perform exponentially weighted smoothing: Current acceleration; Smoothing coefficient; S2. Construct the Kalman filter model and set the state equation and observation equation: The state equation is as follows: The observation equation is: Among them, For the target signal state, To interfere with the input, These are actual observations, including ECG and PGG data. Noise term; S3. Using the Kalman gain correction estimate: S4. Perform outlier identification and processing. If the variation between consecutive peaks is greater than twice the standard deviation, it can be judged as an anomaly.

7. The intelligent wheelchair health monitoring system based on multimodal sensor fusion according to claim 5, characterized in that: The signal processing module also includes a signal reliability screening mechanism, which calculates the wheelchair motion level based on the triaxial acceleration data collected by the IMU sensor, and performs interference removal processing on the physiological signal acquisition results accordingly. The motion level calculation formula is as follows: in , , These represent the instantaneous acceleration values ​​of the wheelchair along the three axes.

8. The intelligent wheelchair health monitoring system based on multimodal sensor fusion according to claim 7, characterized in that: The interference elimination process involves setting a threshold. and stable time ,when Exceeding the set threshold At the same time, the system automatically reduces the weight of PPG and ECG signals in data fusion and blood pressure prediction, and pauses some physiological signal processing to avoid misjudgment caused by motion interference. when Below the threshold When, and the duration exceeds the set stabilization time. At that time, the system initiates dual-modal fusion processing and outputs real-time heart rate and blood pressure values.

9. The intelligent wheelchair health monitoring system based on multimodal sensor fusion according to claim 8, characterized in that: The low-power management module shuts down the high-power ECG sensor when it detects wheelchair movement, keeping only the IMU and PPG working to save power.

10. The intelligent wheelchair health monitoring system based on multimodal sensor fusion according to claim 1, characterized in that: The abnormal states include heart rate exceeding a set threshold, blood pressure fluctuating abnormally, or signal disconnection. The system can trigger a local vibration motor to provide a physical alert.

Images